vtc

Scripts for self-testing on hosted GPU services

Getting Started

Start a new hosted instance

Get a command prompt (cloud IDE or ssh)

cd to the attached filesystem (may not be necessary)

Clone this repo

cd to the repo (vtc)

Install conda

run the new_instance.sh script

run the command 'source /opt/conda/etc/profile.d/conda.sh'

Install burn model

Run the get_burns.sh script

To classifiy some images:

run the command 'conda activate burns_vtc'

cd to the 'Burn-Detection-Classification' directory

run the command 'python detect.py --weights skin_burn_2022_8_21.pt --source YOUR_VIDEO.mp4'

Install pyVHR

cd to the pyvhr directory in this repo

run pyvhr_vtc.sh

To classify some images:

run the command 'conda activate pyvhr_vtc'

edit the videoFileName variable in testRun.py

run the command 'python testRun.py'

MLFlow

The mlflow package is installed via pip in both the VHR and burn startup scripts. You'll still need to start the server if you want to log anything: https://mlflow.org/docs/latest/getting-started/intro-quickstart/index.html#step-2-start-a-tracking-server

import-export

Since we lose all data when shutting down an instance (unless we pay for a filesystem), we need to export experiments from the instance's mlflow server then import them into our ARA mlflow server. The mlflow import-export tool is installed in both vhr and burn environments. Export experiments with: https://github.com/mlflow/mlflow-export-import?tab=readme-ov-file#export-experiment Then download to your local machine